Chris Choy 3D vision research
Index / A Site record

Chris Choy / Spatial index

About

Research background, selected work, education, and contact.

I’m a Senior Research Scientist at NVIDIA, where I work on 3D computer vision, scene understanding, and geometric deep learning. This page collects the research threads, projects, and background behind the shorter bios on my papers and CV.

What I work on

My research sits at the intersection of three things: how to represent 3D space efficiently (sparse tensors, high-dimensional convolutions), how to learn priors over geometry from data (foundation models for 3D, point-cloud transformers), and how to push those models into real systems (robotics, AR/VR, autonomous platforms). My PhD thesis at Stanford, High-dimensional Convolutional Neural Networks for 3D Perception, laid the groundwork for a lot of this — the full thesis is on the Stanford Digital Repository.

A non-exhaustive list of things I’ve spent serious time thinking about: 4D spatio-temporal segmentation, sparse 3D convolutions, point-cloud registration, open-vocabulary 3D understanding, rotation representations, and the strange optimization landscapes that show up when you do gradient descent on manifolds.

Selected publications

I’ve authored or co-authored 53 peer-reviewed papers, with 15K+ citations on Google Scholar. A few that I keep coming back to:

  • MinkowskiNet — 4D spatio-temporal ConvNets. CVPR 2019. Won the 2019 ScanNet semantic-segmentation challenge.
  • Mosaic3D — foundation dataset and model for open-vocabulary 3D segmentation.
  • SpaCeFormer — real-time open-vocabulary 3D instance segmentation without proposals.
  • FCGF — fully convolutional geometric features for point-cloud registration.
  • Universal Correspondence Network — early work on learning correspondences end-to-end.

The full list, with bibtex, is on Google Scholar.

Education

Stanford University

Ph.D. in Electrical Engineering. Thesis: High-dimensional Convolutional Neural Networks for 3D Perception. Advised by Silvio Savarese.

Stanford University

M.S. in Electrical Engineering.

KAIST

B.S. in Electrical Engineering, summa cum laude. Korea Advanced Institute of Science and Technology.

Honors

  • 1st place, 2019 ScanNet Semantic Segmentation Challenge
  • Stanford SystemX FMA Fellowship
  • Korea Foundation for Advanced Studies, Ph.D. Fellowship
  • 5th Presidential Science Fellowship for Undergraduate Study
  • KAIST EECS Merit Scholarship for Best Performance
  • KAIST Alumni Chairman’s Award
  • Korea Science Olympiad, Junior High, Gold Medal

Patents

  • Universal correspondence network. US Patent App. 10/115,032.
  • Systems and methods for semantic segmentation of 3D point clouds. US Patent App. 16/155,843.
  • Computer-based techniques for learning compositional representations of 3D point clouds. US Patent 11,869,149.
  • Action-conditional implicit dynamics of deformable objects. US Patent 12,165,258.
  • Using neural networks to perform object detection, instance segmentation, and semantic correspondence from bounding box supervision. US Patent App. 17/177,068.
  • Scene reconstruction from monocular video. US Patent App. 18/524,803.
  • Sparse voxel transformer for camera-based 3D semantic scene completion. US Patent App. 18/515,016.

Why I write

I write notes when I’m working through something difficult, then realize months later I want to look the same thing up. This site is the indexed version of that habit. If a derivation, a debugging story, or a literature roundup here saves you an afternoon, it has paid for itself.

Contact